With the deep integration of edge computing and Internet of Things (IoT) technologies, the computational capabilities of intelligent edge cameras continue to advance, providing new opportunities for the local deployment of video understanding algorithms. However, existing video captioning models suffer from high computational complexity and large parameter counts, making them challenging to meet the real-time processing requirements of resource-constrained IoT edge devices. In this work, we propose EdgeVidCap, a lightweight video captioning model specifically designed for IoT edge cameras. Specifically, we design a hybrid module termed Synergetic Attention State Mamba (SASM) that incorporates channel attention mechanisms to enhance feature selection capabilities and leverages State Space Models (SSMs) to efficiently capture long-range spatial dependencies, achieving efficient spatiotemporal modeling of multimodal video features. In the caption generation stage, we propose an adaptive attention-guided LSTM decoder that can dynamically adjust feature weights according to video content and auto-regressively generate semantically rich and accurate textual descriptions. Comprehensive evaluations of EdgeVidCap on mainstream datasets, including MSR-VTT and MSVD are analyzed. Experimental results demonstrate that our system demonstrated enhanced precision relative to existing investigations, and our streamlined frame filtering mechanism yielded greater processing efficiency while creating more dependable descriptions following frame selection.
Few-Shot Semantic Segmentation (FSS) faces significant challenges in modeling complex backgrounds and maintaining prediction consistency due to limited training samples. Existing methods oversimplify backgrounds as single negative classes and rely solely on pixel-level alignments. To address these issues, we propose a symmetry-aware superpixel-enhanced FSS framework with a symmetric dual-branch architecture that explicitly models the superpixel region-graph in both the support and query branches. First, top–down cross-layer fusion injects low-level edge and texture cues into high-level semantics to build a more complete representation of complex backgrounds, improving foreground–background separability and boundary quality. Second, images are partitioned into superpixels and aggregated into “superpixel tokens” to construct a Region Adjacency Graph (RAG). Support-set prototypes are used to initialize query-pixel predictions, which are then projected into the superpixel space for cross-image prototype alignment with support superpixels. We further perform message passing/energy minimization on the RAG to enhance intra-region consistency and boundary adherence, and finally back-project the predictions to the pixel space. Lastly, by aggregating homogeneous semantic information, we construct robust foreground and background prototype representations, enhancing the model’s ability to perceive both seen and novel targets. Extensive experiments on the PASCAL- and COCO- benchmarks demonstrate that our proposed model achieves superior segmentation performance over the baseline and remains competitive with existing FSS methods.